1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/slab.h>
7 #include <linux/blkdev.h>
8 #include <linux/writeback.h>
9 #include <linux/sched/mm.h>
13 #include "transaction.h"
14 #include "btrfs_inode.h"
15 #include "extent_io.h"
17 #include "compression.h"
18 #include "delalloc-space.h"
24 static struct kmem_cache *btrfs_ordered_extent_cache;
26 static u64 entry_end(struct btrfs_ordered_extent *entry)
28 if (entry->file_offset + entry->num_bytes < entry->file_offset)
30 return entry->file_offset + entry->num_bytes;
33 /* returns NULL if the insertion worked, or it returns the node it did find
36 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
39 struct rb_node **p = &root->rb_node;
40 struct rb_node *parent = NULL;
41 struct btrfs_ordered_extent *entry;
45 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
47 if (file_offset < entry->file_offset)
49 else if (file_offset >= entry_end(entry))
55 rb_link_node(node, parent, p);
56 rb_insert_color(node, root);
61 * look for a given offset in the tree, and if it can't be found return the
64 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
65 struct rb_node **prev_ret)
67 struct rb_node *n = root->rb_node;
68 struct rb_node *prev = NULL;
70 struct btrfs_ordered_extent *entry;
71 struct btrfs_ordered_extent *prev_entry = NULL;
74 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
78 if (file_offset < entry->file_offset)
80 else if (file_offset >= entry_end(entry))
88 while (prev && file_offset >= entry_end(prev_entry)) {
92 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
94 if (file_offset < entry_end(prev_entry))
100 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
102 while (prev && file_offset < entry_end(prev_entry)) {
103 test = rb_prev(prev);
106 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
114 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
117 if (file_offset + len <= entry->file_offset ||
118 entry->file_offset + entry->num_bytes <= file_offset)
124 * look find the first ordered struct that has this offset, otherwise
125 * the first one less than this offset
127 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
130 struct rb_root *root = &tree->tree;
131 struct rb_node *prev = NULL;
133 struct btrfs_ordered_extent *entry;
136 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
138 if (in_range(file_offset, entry->file_offset, entry->num_bytes))
141 ret = __tree_search(root, file_offset, &prev);
149 static struct btrfs_ordered_extent *alloc_ordered_extent(
150 struct btrfs_inode *inode, u64 file_offset, u64 num_bytes,
151 u64 ram_bytes, u64 disk_bytenr, u64 disk_num_bytes,
152 u64 offset, unsigned long flags, int compress_type)
154 struct btrfs_ordered_extent *entry;
158 ((1 << BTRFS_ORDERED_NOCOW) | (1 << BTRFS_ORDERED_PREALLOC))) {
159 /* For nocow write, we can release the qgroup rsv right now */
160 ret = btrfs_qgroup_free_data(inode, NULL, file_offset, num_bytes);
165 * The ordered extent has reserved qgroup space, release now
166 * and pass the reserved number for qgroup_record to free.
168 ret = btrfs_qgroup_release_data(inode, file_offset, num_bytes);
172 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
174 return ERR_PTR(-ENOMEM);
176 entry->file_offset = file_offset;
177 entry->num_bytes = num_bytes;
178 entry->ram_bytes = ram_bytes;
179 entry->disk_bytenr = disk_bytenr;
180 entry->disk_num_bytes = disk_num_bytes;
181 entry->offset = offset;
182 entry->bytes_left = num_bytes;
183 entry->inode = igrab(&inode->vfs_inode);
184 entry->compress_type = compress_type;
185 entry->truncated_len = (u64)-1;
186 entry->qgroup_rsv = ret;
187 entry->flags = flags;
188 refcount_set(&entry->refs, 1);
189 init_waitqueue_head(&entry->wait);
190 INIT_LIST_HEAD(&entry->list);
191 INIT_LIST_HEAD(&entry->log_list);
192 INIT_LIST_HEAD(&entry->root_extent_list);
193 INIT_LIST_HEAD(&entry->work_list);
194 init_completion(&entry->completion);
197 * We don't need the count_max_extents here, we can assume that all of
198 * that work has been done at higher layers, so this is truly the
199 * smallest the extent is going to get.
201 spin_lock(&inode->lock);
202 btrfs_mod_outstanding_extents(inode, 1);
203 spin_unlock(&inode->lock);
208 static void insert_ordered_extent(struct btrfs_ordered_extent *entry)
210 struct btrfs_inode *inode = BTRFS_I(entry->inode);
211 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
212 struct btrfs_root *root = inode->root;
213 struct btrfs_fs_info *fs_info = root->fs_info;
214 struct rb_node *node;
216 trace_btrfs_ordered_extent_add(inode, entry);
218 percpu_counter_add_batch(&fs_info->ordered_bytes, entry->num_bytes,
219 fs_info->delalloc_batch);
221 /* One ref for the tree. */
222 refcount_inc(&entry->refs);
224 spin_lock_irq(&tree->lock);
225 node = tree_insert(&tree->tree, entry->file_offset, &entry->rb_node);
227 btrfs_panic(fs_info, -EEXIST,
228 "inconsistency in ordered tree at offset %llu",
230 spin_unlock_irq(&tree->lock);
232 spin_lock(&root->ordered_extent_lock);
233 list_add_tail(&entry->root_extent_list,
234 &root->ordered_extents);
235 root->nr_ordered_extents++;
236 if (root->nr_ordered_extents == 1) {
237 spin_lock(&fs_info->ordered_root_lock);
238 BUG_ON(!list_empty(&root->ordered_root));
239 list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
240 spin_unlock(&fs_info->ordered_root_lock);
242 spin_unlock(&root->ordered_extent_lock);
246 * Add an ordered extent to the per-inode tree.
248 * @inode: Inode that this extent is for.
249 * @file_offset: Logical offset in file where the extent starts.
250 * @num_bytes: Logical length of extent in file.
251 * @ram_bytes: Full length of unencoded data.
252 * @disk_bytenr: Offset of extent on disk.
253 * @disk_num_bytes: Size of extent on disk.
254 * @offset: Offset into unencoded data where file data starts.
255 * @flags: Flags specifying type of extent (1 << BTRFS_ORDERED_*).
256 * @compress_type: Compression algorithm used for data.
258 * Most of these parameters correspond to &struct btrfs_file_extent_item. The
259 * tree is given a single reference on the ordered extent that was inserted, and
260 * the returned pointer is given a second reference.
262 * Return: the new ordered extent or error pointer.
264 struct btrfs_ordered_extent *btrfs_alloc_ordered_extent(
265 struct btrfs_inode *inode, u64 file_offset,
266 u64 num_bytes, u64 ram_bytes, u64 disk_bytenr,
267 u64 disk_num_bytes, u64 offset, unsigned long flags,
270 struct btrfs_ordered_extent *entry;
272 ASSERT((flags & ~BTRFS_ORDERED_TYPE_FLAGS) == 0);
274 entry = alloc_ordered_extent(inode, file_offset, num_bytes, ram_bytes,
275 disk_bytenr, disk_num_bytes, offset, flags,
278 insert_ordered_extent(entry);
283 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
284 * when an ordered extent is finished. If the list covers more than one
285 * ordered extent, it is split across multiples.
287 void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry,
288 struct btrfs_ordered_sum *sum)
290 struct btrfs_ordered_inode_tree *tree;
292 tree = &BTRFS_I(entry->inode)->ordered_tree;
293 spin_lock_irq(&tree->lock);
294 list_add_tail(&sum->list, &entry->list);
295 spin_unlock_irq(&tree->lock);
298 static void finish_ordered_fn(struct btrfs_work *work)
300 struct btrfs_ordered_extent *ordered_extent;
302 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
303 btrfs_finish_ordered_io(ordered_extent);
306 static bool can_finish_ordered_extent(struct btrfs_ordered_extent *ordered,
307 struct page *page, u64 file_offset,
308 u64 len, bool uptodate)
310 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
311 struct btrfs_fs_info *fs_info = inode->root->fs_info;
313 lockdep_assert_held(&inode->ordered_tree.lock);
316 ASSERT(page->mapping);
317 ASSERT(page_offset(page) <= file_offset);
318 ASSERT(file_offset + len <= page_offset(page) + PAGE_SIZE);
321 * Ordered (Private2) bit indicates whether we still have
322 * pending io unfinished for the ordered extent.
324 * If there's no such bit, we need to skip to next range.
326 if (!btrfs_page_test_ordered(fs_info, page, file_offset, len))
328 btrfs_page_clear_ordered(fs_info, page, file_offset, len);
331 /* Now we're fine to update the accounting. */
332 if (WARN_ON_ONCE(len > ordered->bytes_left)) {
334 "bad ordered extent accounting, root=%llu ino=%llu OE offset=%llu OE len=%llu to_dec=%llu left=%llu",
335 inode->root->root_key.objectid, btrfs_ino(inode),
336 ordered->file_offset, ordered->num_bytes,
337 len, ordered->bytes_left);
338 ordered->bytes_left = 0;
340 ordered->bytes_left -= len;
344 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
346 if (ordered->bytes_left)
350 * All the IO of the ordered extent is finished, we need to queue
351 * the finish_func to be executed.
353 set_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags);
354 cond_wake_up(&ordered->wait);
355 refcount_inc(&ordered->refs);
356 trace_btrfs_ordered_extent_mark_finished(inode, ordered);
360 static void btrfs_queue_ordered_fn(struct btrfs_ordered_extent *ordered)
362 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
363 struct btrfs_fs_info *fs_info = inode->root->fs_info;
364 struct btrfs_workqueue *wq = btrfs_is_free_space_inode(inode) ?
365 fs_info->endio_freespace_worker : fs_info->endio_write_workers;
367 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL, NULL);
368 btrfs_queue_work(wq, &ordered->work);
371 bool btrfs_finish_ordered_extent(struct btrfs_ordered_extent *ordered,
372 struct page *page, u64 file_offset, u64 len,
375 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
379 trace_btrfs_finish_ordered_extent(inode, file_offset, len, uptodate);
381 spin_lock_irqsave(&inode->ordered_tree.lock, flags);
382 ret = can_finish_ordered_extent(ordered, page, file_offset, len, uptodate);
383 spin_unlock_irqrestore(&inode->ordered_tree.lock, flags);
386 btrfs_queue_ordered_fn(ordered);
391 * Mark all ordered extents io inside the specified range finished.
393 * @page: The involved page for the operation.
394 * For uncompressed buffered IO, the page status also needs to be
395 * updated to indicate whether the pending ordered io is finished.
396 * Can be NULL for direct IO and compressed write.
397 * For these cases, callers are ensured they won't execute the
398 * endio function twice.
400 * This function is called for endio, thus the range must have ordered
401 * extent(s) covering it.
403 void btrfs_mark_ordered_io_finished(struct btrfs_inode *inode,
404 struct page *page, u64 file_offset,
405 u64 num_bytes, bool uptodate)
407 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
408 struct rb_node *node;
409 struct btrfs_ordered_extent *entry = NULL;
411 u64 cur = file_offset;
413 trace_btrfs_writepage_end_io_hook(inode, file_offset,
414 file_offset + num_bytes - 1,
417 spin_lock_irqsave(&tree->lock, flags);
418 while (cur < file_offset + num_bytes) {
423 node = tree_search(tree, cur);
424 /* No ordered extents at all */
428 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
429 entry_end = entry->file_offset + entry->num_bytes;
435 if (cur >= entry_end) {
436 node = rb_next(node);
437 /* No more ordered extents, exit */
440 entry = rb_entry(node, struct btrfs_ordered_extent,
443 /* Go to next ordered extent and continue */
444 cur = entry->file_offset;
450 * Go to the start of OE.
452 if (cur < entry->file_offset) {
453 cur = entry->file_offset;
458 * Now we are definitely inside one ordered extent.
464 end = min(entry->file_offset + entry->num_bytes,
465 file_offset + num_bytes) - 1;
466 ASSERT(end + 1 - cur < U32_MAX);
469 if (can_finish_ordered_extent(entry, page, cur, len, uptodate)) {
470 spin_unlock_irqrestore(&tree->lock, flags);
471 btrfs_queue_ordered_fn(entry);
472 spin_lock_irqsave(&tree->lock, flags);
476 spin_unlock_irqrestore(&tree->lock, flags);
480 * Finish IO for one ordered extent across a given range. The range can only
481 * contain one ordered extent.
483 * @cached: The cached ordered extent. If not NULL, we can skip the tree
484 * search and use the ordered extent directly.
485 * Will be also used to store the finished ordered extent.
486 * @file_offset: File offset for the finished IO
487 * @io_size: Length of the finish IO range
489 * Return true if the ordered extent is finished in the range, and update
491 * Return false otherwise.
493 * NOTE: The range can NOT cross multiple ordered extents.
494 * Thus caller should ensure the range doesn't cross ordered extents.
496 bool btrfs_dec_test_ordered_pending(struct btrfs_inode *inode,
497 struct btrfs_ordered_extent **cached,
498 u64 file_offset, u64 io_size)
500 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
501 struct rb_node *node;
502 struct btrfs_ordered_extent *entry = NULL;
504 bool finished = false;
506 spin_lock_irqsave(&tree->lock, flags);
507 if (cached && *cached) {
512 node = tree_search(tree, file_offset);
516 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
518 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
521 if (io_size > entry->bytes_left)
522 btrfs_crit(inode->root->fs_info,
523 "bad ordered accounting left %llu size %llu",
524 entry->bytes_left, io_size);
526 entry->bytes_left -= io_size;
528 if (entry->bytes_left == 0) {
530 * Ensure only one caller can set the flag and finished_ret
533 finished = !test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
534 /* test_and_set_bit implies a barrier */
535 cond_wake_up_nomb(&entry->wait);
538 if (finished && cached && entry) {
540 refcount_inc(&entry->refs);
541 trace_btrfs_ordered_extent_dec_test_pending(inode, entry);
543 spin_unlock_irqrestore(&tree->lock, flags);
548 * used to drop a reference on an ordered extent. This will free
549 * the extent if the last reference is dropped
551 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
553 struct list_head *cur;
554 struct btrfs_ordered_sum *sum;
556 trace_btrfs_ordered_extent_put(BTRFS_I(entry->inode), entry);
558 if (refcount_dec_and_test(&entry->refs)) {
559 ASSERT(list_empty(&entry->root_extent_list));
560 ASSERT(list_empty(&entry->log_list));
561 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
563 btrfs_add_delayed_iput(BTRFS_I(entry->inode));
564 while (!list_empty(&entry->list)) {
565 cur = entry->list.next;
566 sum = list_entry(cur, struct btrfs_ordered_sum, list);
567 list_del(&sum->list);
570 kmem_cache_free(btrfs_ordered_extent_cache, entry);
575 * remove an ordered extent from the tree. No references are dropped
576 * and waiters are woken up.
578 void btrfs_remove_ordered_extent(struct btrfs_inode *btrfs_inode,
579 struct btrfs_ordered_extent *entry)
581 struct btrfs_ordered_inode_tree *tree;
582 struct btrfs_root *root = btrfs_inode->root;
583 struct btrfs_fs_info *fs_info = root->fs_info;
584 struct rb_node *node;
586 bool freespace_inode;
589 * If this is a free space inode the thread has not acquired the ordered
590 * extents lockdep map.
592 freespace_inode = btrfs_is_free_space_inode(btrfs_inode);
594 btrfs_lockdep_acquire(fs_info, btrfs_trans_pending_ordered);
595 /* This is paired with btrfs_alloc_ordered_extent. */
596 spin_lock(&btrfs_inode->lock);
597 btrfs_mod_outstanding_extents(btrfs_inode, -1);
598 spin_unlock(&btrfs_inode->lock);
599 if (root != fs_info->tree_root) {
602 if (test_bit(BTRFS_ORDERED_ENCODED, &entry->flags))
603 release = entry->disk_num_bytes;
605 release = entry->num_bytes;
606 btrfs_delalloc_release_metadata(btrfs_inode, release,
607 test_bit(BTRFS_ORDERED_IOERR,
611 percpu_counter_add_batch(&fs_info->ordered_bytes, -entry->num_bytes,
612 fs_info->delalloc_batch);
614 tree = &btrfs_inode->ordered_tree;
615 spin_lock_irq(&tree->lock);
616 node = &entry->rb_node;
617 rb_erase(node, &tree->tree);
619 if (tree->last == node)
621 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
622 pending = test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags);
623 spin_unlock_irq(&tree->lock);
626 * The current running transaction is waiting on us, we need to let it
627 * know that we're complete and wake it up.
630 struct btrfs_transaction *trans;
633 * The checks for trans are just a formality, it should be set,
634 * but if it isn't we don't want to deref/assert under the spin
635 * lock, so be nice and check if trans is set, but ASSERT() so
636 * if it isn't set a developer will notice.
638 spin_lock(&fs_info->trans_lock);
639 trans = fs_info->running_transaction;
641 refcount_inc(&trans->use_count);
642 spin_unlock(&fs_info->trans_lock);
644 ASSERT(trans || BTRFS_FS_ERROR(fs_info));
646 if (atomic_dec_and_test(&trans->pending_ordered))
647 wake_up(&trans->pending_wait);
648 btrfs_put_transaction(trans);
652 btrfs_lockdep_release(fs_info, btrfs_trans_pending_ordered);
654 spin_lock(&root->ordered_extent_lock);
655 list_del_init(&entry->root_extent_list);
656 root->nr_ordered_extents--;
658 trace_btrfs_ordered_extent_remove(btrfs_inode, entry);
660 if (!root->nr_ordered_extents) {
661 spin_lock(&fs_info->ordered_root_lock);
662 BUG_ON(list_empty(&root->ordered_root));
663 list_del_init(&root->ordered_root);
664 spin_unlock(&fs_info->ordered_root_lock);
666 spin_unlock(&root->ordered_extent_lock);
667 wake_up(&entry->wait);
668 if (!freespace_inode)
669 btrfs_lockdep_release(fs_info, btrfs_ordered_extent);
672 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
674 struct btrfs_ordered_extent *ordered;
676 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
677 btrfs_start_ordered_extent(ordered);
678 complete(&ordered->completion);
682 * wait for all the ordered extents in a root. This is done when balancing
683 * space between drives.
685 u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
686 const u64 range_start, const u64 range_len)
688 struct btrfs_fs_info *fs_info = root->fs_info;
692 struct btrfs_ordered_extent *ordered, *next;
694 const u64 range_end = range_start + range_len;
696 mutex_lock(&root->ordered_extent_mutex);
697 spin_lock(&root->ordered_extent_lock);
698 list_splice_init(&root->ordered_extents, &splice);
699 while (!list_empty(&splice) && nr) {
700 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
703 if (range_end <= ordered->disk_bytenr ||
704 ordered->disk_bytenr + ordered->disk_num_bytes <= range_start) {
705 list_move_tail(&ordered->root_extent_list, &skipped);
706 cond_resched_lock(&root->ordered_extent_lock);
710 list_move_tail(&ordered->root_extent_list,
711 &root->ordered_extents);
712 refcount_inc(&ordered->refs);
713 spin_unlock(&root->ordered_extent_lock);
715 btrfs_init_work(&ordered->flush_work,
716 btrfs_run_ordered_extent_work, NULL, NULL);
717 list_add_tail(&ordered->work_list, &works);
718 btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
721 spin_lock(&root->ordered_extent_lock);
726 list_splice_tail(&skipped, &root->ordered_extents);
727 list_splice_tail(&splice, &root->ordered_extents);
728 spin_unlock(&root->ordered_extent_lock);
730 list_for_each_entry_safe(ordered, next, &works, work_list) {
731 list_del_init(&ordered->work_list);
732 wait_for_completion(&ordered->completion);
733 btrfs_put_ordered_extent(ordered);
736 mutex_unlock(&root->ordered_extent_mutex);
741 void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
742 const u64 range_start, const u64 range_len)
744 struct btrfs_root *root;
748 mutex_lock(&fs_info->ordered_operations_mutex);
749 spin_lock(&fs_info->ordered_root_lock);
750 list_splice_init(&fs_info->ordered_roots, &splice);
751 while (!list_empty(&splice) && nr) {
752 root = list_first_entry(&splice, struct btrfs_root,
754 root = btrfs_grab_root(root);
756 list_move_tail(&root->ordered_root,
757 &fs_info->ordered_roots);
758 spin_unlock(&fs_info->ordered_root_lock);
760 done = btrfs_wait_ordered_extents(root, nr,
761 range_start, range_len);
762 btrfs_put_root(root);
764 spin_lock(&fs_info->ordered_root_lock);
769 list_splice_tail(&splice, &fs_info->ordered_roots);
770 spin_unlock(&fs_info->ordered_root_lock);
771 mutex_unlock(&fs_info->ordered_operations_mutex);
775 * Start IO and wait for a given ordered extent to finish.
777 * Wait on page writeback for all the pages in the extent and the IO completion
778 * code to insert metadata into the btree corresponding to the extent.
780 void btrfs_start_ordered_extent(struct btrfs_ordered_extent *entry)
782 u64 start = entry->file_offset;
783 u64 end = start + entry->num_bytes - 1;
784 struct btrfs_inode *inode = BTRFS_I(entry->inode);
785 bool freespace_inode;
787 trace_btrfs_ordered_extent_start(inode, entry);
790 * If this is a free space inode do not take the ordered extents lockdep
793 freespace_inode = btrfs_is_free_space_inode(inode);
796 * pages in the range can be dirty, clean or writeback. We
797 * start IO on any dirty ones so the wait doesn't stall waiting
798 * for the flusher thread to find them
800 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
801 filemap_fdatawrite_range(inode->vfs_inode.i_mapping, start, end);
803 if (!freespace_inode)
804 btrfs_might_wait_for_event(inode->root->fs_info, btrfs_ordered_extent);
805 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, &entry->flags));
809 * Used to wait on ordered extents across a large range of bytes.
811 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
817 struct btrfs_ordered_extent *ordered;
819 if (start + len < start) {
820 orig_end = OFFSET_MAX;
822 orig_end = start + len - 1;
823 if (orig_end > OFFSET_MAX)
824 orig_end = OFFSET_MAX;
827 /* start IO across the range first to instantiate any delalloc
830 ret = btrfs_fdatawrite_range(inode, start, orig_end);
835 * If we have a writeback error don't return immediately. Wait first
836 * for any ordered extents that haven't completed yet. This is to make
837 * sure no one can dirty the same page ranges and call writepages()
838 * before the ordered extents complete - to avoid failures (-EEXIST)
839 * when adding the new ordered extents to the ordered tree.
841 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
845 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode), end);
848 if (ordered->file_offset > orig_end) {
849 btrfs_put_ordered_extent(ordered);
852 if (ordered->file_offset + ordered->num_bytes <= start) {
853 btrfs_put_ordered_extent(ordered);
856 btrfs_start_ordered_extent(ordered);
857 end = ordered->file_offset;
859 * If the ordered extent had an error save the error but don't
860 * exit without waiting first for all other ordered extents in
861 * the range to complete.
863 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
865 btrfs_put_ordered_extent(ordered);
866 if (end == 0 || end == start)
870 return ret_wb ? ret_wb : ret;
874 * find an ordered extent corresponding to file_offset. return NULL if
875 * nothing is found, otherwise take a reference on the extent and return it
877 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct btrfs_inode *inode,
880 struct btrfs_ordered_inode_tree *tree;
881 struct rb_node *node;
882 struct btrfs_ordered_extent *entry = NULL;
885 tree = &inode->ordered_tree;
886 spin_lock_irqsave(&tree->lock, flags);
887 node = tree_search(tree, file_offset);
891 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
892 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
895 refcount_inc(&entry->refs);
896 trace_btrfs_ordered_extent_lookup(inode, entry);
899 spin_unlock_irqrestore(&tree->lock, flags);
903 /* Since the DIO code tries to lock a wide area we need to look for any ordered
904 * extents that exist in the range, rather than just the start of the range.
906 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
907 struct btrfs_inode *inode, u64 file_offset, u64 len)
909 struct btrfs_ordered_inode_tree *tree;
910 struct rb_node *node;
911 struct btrfs_ordered_extent *entry = NULL;
913 tree = &inode->ordered_tree;
914 spin_lock_irq(&tree->lock);
915 node = tree_search(tree, file_offset);
917 node = tree_search(tree, file_offset + len);
923 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
924 if (range_overlaps(entry, file_offset, len))
927 if (entry->file_offset >= file_offset + len) {
932 node = rb_next(node);
938 refcount_inc(&entry->refs);
939 trace_btrfs_ordered_extent_lookup_range(inode, entry);
941 spin_unlock_irq(&tree->lock);
946 * Adds all ordered extents to the given list. The list ends up sorted by the
947 * file_offset of the ordered extents.
949 void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode,
950 struct list_head *list)
952 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
955 ASSERT(inode_is_locked(&inode->vfs_inode));
957 spin_lock_irq(&tree->lock);
958 for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
959 struct btrfs_ordered_extent *ordered;
961 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
963 if (test_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
966 ASSERT(list_empty(&ordered->log_list));
967 list_add_tail(&ordered->log_list, list);
968 refcount_inc(&ordered->refs);
969 trace_btrfs_ordered_extent_lookup_for_logging(inode, ordered);
971 spin_unlock_irq(&tree->lock);
975 * lookup and return any extent before 'file_offset'. NULL is returned
978 struct btrfs_ordered_extent *
979 btrfs_lookup_first_ordered_extent(struct btrfs_inode *inode, u64 file_offset)
981 struct btrfs_ordered_inode_tree *tree;
982 struct rb_node *node;
983 struct btrfs_ordered_extent *entry = NULL;
985 tree = &inode->ordered_tree;
986 spin_lock_irq(&tree->lock);
987 node = tree_search(tree, file_offset);
991 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
992 refcount_inc(&entry->refs);
993 trace_btrfs_ordered_extent_lookup_first(inode, entry);
995 spin_unlock_irq(&tree->lock);
1000 * Lookup the first ordered extent that overlaps the range
1001 * [@file_offset, @file_offset + @len).
1003 * The difference between this and btrfs_lookup_first_ordered_extent() is
1004 * that this one won't return any ordered extent that does not overlap the range.
1005 * And the difference against btrfs_lookup_ordered_extent() is, this function
1006 * ensures the first ordered extent gets returned.
1008 struct btrfs_ordered_extent *btrfs_lookup_first_ordered_range(
1009 struct btrfs_inode *inode, u64 file_offset, u64 len)
1011 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
1012 struct rb_node *node;
1013 struct rb_node *cur;
1014 struct rb_node *prev;
1015 struct rb_node *next;
1016 struct btrfs_ordered_extent *entry = NULL;
1018 spin_lock_irq(&tree->lock);
1019 node = tree->tree.rb_node;
1021 * Here we don't want to use tree_search() which will use tree->last
1022 * and screw up the search order.
1023 * And __tree_search() can't return the adjacent ordered extents
1024 * either, thus here we do our own search.
1027 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
1029 if (file_offset < entry->file_offset) {
1030 node = node->rb_left;
1031 } else if (file_offset >= entry_end(entry)) {
1032 node = node->rb_right;
1035 * Direct hit, got an ordered extent that starts at
1046 cur = &entry->rb_node;
1047 /* We got an entry around @file_offset, check adjacent entries */
1048 if (entry->file_offset < file_offset) {
1050 next = rb_next(cur);
1052 prev = rb_prev(cur);
1056 entry = rb_entry(prev, struct btrfs_ordered_extent, rb_node);
1057 if (range_overlaps(entry, file_offset, len))
1061 entry = rb_entry(next, struct btrfs_ordered_extent, rb_node);
1062 if (range_overlaps(entry, file_offset, len))
1065 /* No ordered extent in the range */
1069 refcount_inc(&entry->refs);
1070 trace_btrfs_ordered_extent_lookup_first_range(inode, entry);
1073 spin_unlock_irq(&tree->lock);
1078 * Lock the passed range and ensures all pending ordered extents in it are run
1081 * @inode: Inode whose ordered tree is to be searched
1082 * @start: Beginning of range to flush
1083 * @end: Last byte of range to lock
1084 * @cached_state: If passed, will return the extent state responsible for the
1085 * locked range. It's the caller's responsibility to free the
1088 * Always return with the given range locked, ensuring after it's called no
1089 * order extent can be pending.
1091 void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start,
1093 struct extent_state **cached_state)
1095 struct btrfs_ordered_extent *ordered;
1096 struct extent_state *cache = NULL;
1097 struct extent_state **cachedp = &cache;
1100 cachedp = cached_state;
1103 lock_extent(&inode->io_tree, start, end, cachedp);
1104 ordered = btrfs_lookup_ordered_range(inode, start,
1108 * If no external cached_state has been passed then
1109 * decrement the extra ref taken for cachedp since we
1110 * aren't exposing it outside of this function
1113 refcount_dec(&cache->refs);
1116 unlock_extent(&inode->io_tree, start, end, cachedp);
1117 btrfs_start_ordered_extent(ordered);
1118 btrfs_put_ordered_extent(ordered);
1123 * Lock the passed range and ensure all pending ordered extents in it are run
1124 * to completion in nowait mode.
1126 * Return true if btrfs_lock_ordered_range does not return any extents,
1129 bool btrfs_try_lock_ordered_range(struct btrfs_inode *inode, u64 start, u64 end,
1130 struct extent_state **cached_state)
1132 struct btrfs_ordered_extent *ordered;
1134 if (!try_lock_extent(&inode->io_tree, start, end, cached_state))
1137 ordered = btrfs_lookup_ordered_range(inode, start, end - start + 1);
1141 btrfs_put_ordered_extent(ordered);
1142 unlock_extent(&inode->io_tree, start, end, cached_state);
1147 /* Split out a new ordered extent for this first @len bytes of @ordered. */
1148 struct btrfs_ordered_extent *btrfs_split_ordered_extent(
1149 struct btrfs_ordered_extent *ordered, u64 len)
1151 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1152 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
1153 struct btrfs_root *root = inode->root;
1154 struct btrfs_fs_info *fs_info = root->fs_info;
1155 u64 file_offset = ordered->file_offset;
1156 u64 disk_bytenr = ordered->disk_bytenr;
1157 unsigned long flags = ordered->flags;
1158 struct btrfs_ordered_sum *sum, *tmpsum;
1159 struct btrfs_ordered_extent *new;
1160 struct rb_node *node;
1163 trace_btrfs_ordered_extent_split(inode, ordered);
1165 ASSERT(!(flags & (1U << BTRFS_ORDERED_COMPRESSED)));
1168 * The entire bio must be covered by the ordered extent, but we can't
1169 * reduce the original extent to a zero length either.
1171 if (WARN_ON_ONCE(len >= ordered->num_bytes))
1172 return ERR_PTR(-EINVAL);
1173 /* We cannot split partially completed ordered extents. */
1174 if (ordered->bytes_left) {
1175 ASSERT(!(flags & ~BTRFS_ORDERED_TYPE_FLAGS));
1176 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes))
1177 return ERR_PTR(-EINVAL);
1179 /* We cannot split a compressed ordered extent. */
1180 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes))
1181 return ERR_PTR(-EINVAL);
1183 new = alloc_ordered_extent(inode, file_offset, len, len, disk_bytenr,
1184 len, 0, flags, ordered->compress_type);
1188 /* One ref for the tree. */
1189 refcount_inc(&new->refs);
1191 spin_lock_irq(&root->ordered_extent_lock);
1192 spin_lock(&tree->lock);
1193 /* Remove from tree once */
1194 node = &ordered->rb_node;
1195 rb_erase(node, &tree->tree);
1196 RB_CLEAR_NODE(node);
1197 if (tree->last == node)
1200 ordered->file_offset += len;
1201 ordered->disk_bytenr += len;
1202 ordered->num_bytes -= len;
1203 ordered->disk_num_bytes -= len;
1205 if (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags)) {
1206 ASSERT(ordered->bytes_left == 0);
1207 new->bytes_left = 0;
1209 ordered->bytes_left -= len;
1212 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags)) {
1213 if (ordered->truncated_len > len) {
1214 ordered->truncated_len -= len;
1216 new->truncated_len = ordered->truncated_len;
1217 ordered->truncated_len = 0;
1221 list_for_each_entry_safe(sum, tmpsum, &ordered->list, list) {
1224 list_move_tail(&sum->list, &new->list);
1228 /* Re-insert the node */
1229 node = tree_insert(&tree->tree, ordered->file_offset, &ordered->rb_node);
1231 btrfs_panic(fs_info, -EEXIST,
1232 "zoned: inconsistency in ordered tree at offset %llu",
1233 ordered->file_offset);
1235 node = tree_insert(&tree->tree, new->file_offset, &new->rb_node);
1237 btrfs_panic(fs_info, -EEXIST,
1238 "zoned: inconsistency in ordered tree at offset %llu",
1240 spin_unlock(&tree->lock);
1242 list_add_tail(&new->root_extent_list, &root->ordered_extents);
1243 root->nr_ordered_extents++;
1244 spin_unlock_irq(&root->ordered_extent_lock);
1248 int __init ordered_data_init(void)
1250 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1251 sizeof(struct btrfs_ordered_extent), 0,
1254 if (!btrfs_ordered_extent_cache)
1260 void __cold ordered_data_exit(void)
1262 kmem_cache_destroy(btrfs_ordered_extent_cache);